ZFS included in pool with ext4: disk utilization misreporting

[tigerkzr]

I recently made a huge(for me) expansion in drive space with a multi-vdev raidz pool. Since my new space isn't large enough to accommodate all of my data in a way that I could transfer the data and recover my drives as additional zfs space, I have simply added one of the filesystems on the zfs pool to my existing mergerfs pool, all formatted as ext4.

This works great except for the drive space reporting. My mergerfs pool reports only the space utilization on the ext4 portion of the pool. Individually, all drives, and zfs filesystems, report correct utilization data.

Has anyone else encountered this? I'm hesitant to report it as a bug as there might be a setting I am using causing this. here is my mergerfs fstab entry if it helps identify my issue.(line breaks for clarity only)

/mnt/datapool/D*:/mnt/datapool/zdata       /mnt/mergerfs      fuse.mergerfs \

defaults,
fsname=data_pool-nfs,
noforget,
lazy-umount-mountpoint=true,
parallel-direct-writes=true,
process-thread-count=0,
read-thread-count=0,
readahead=5120,
async_read=false,
inodecalc=path-hash,
security_capability=false,
xattr=nosys,
category.create=pfrd,
func.getattr=newest,
cache.files=off,
symlinkify=true  

0 0

the only additional info I have is that the utilization reporting is the same with xattr=passthrough (used while mergerfs.balance is run)

[trapexit]

https://trapexit.github.io/mergerfs/support/

Please follow the instructions in the ticket wrt providing details.

Did you run df as root? Did you confirm permissions are correct? Did you strace mergerfs while running df?

[adhdtiger]

will do... yes. df was run as root. Let me gather the info and I'll get back

[adhdtiger]

Sigh... nevermind. Solved it. Was mounting my zfs filesystem into a container one directory below the mergerfs pool directory to accommodate directory structure. The zfs filesystem was named with the same name as the common diretory the disks share at their root.

so basically I was telling mergerfs to make my merged mount comprised of ext4 filesystem mountpoints plus one directory with a zfs filesystem mountpoint in it.

lxc.datapool/
      disk1.mountpoint/<common dir>
      disk2.mountpoint/<common dir>
      disk3.mountpoint/<common dir>
      ...
      <directory>/zfs.fs.root.mountpoint(named: <common dir>)
 

after moving subdirectories on zfs.fs.root into zfs.fs.root/ instead of relying on the zfs.fs.root being NAMED , changed mountpoints to:

lxc.datapool/
      disk1.mountpoint/<common dir>
      disk2.mountpoint/<common dir>
      disk3.mountpoint/<common dir>
      ...
      zfs.fs.root.mountpoint/<common dir>

and now utilization is reported with the proper sizes.

ETA: these were host mountpoints bind mounted into an LXC container running mergerfs

[trapexit]

Yeah, mergerfs does not try to traverse the tree nor would it make much sense to aggregate the values since it's not that simple. It does offer some settings for how to do statfs.